CALTECH CS137 Winter DeHon CS137: Electronic Design Automation Day 7: February 3, 2002 Retiming
CALTECH CS137 Winter DeHon Today Retiming –Cycle time (clock period) –C-slow –Initial states –Register minimization –Necessary delays (time permitting)
CALTECH CS137 Winter DeHon Task Move registers to: –Preserve semantics –Minimize path length between registers –(make path length 1 for maximum throughput or reuse) –Maximize reuse rate –…while minimizing number of registers required
CALTECH CS137 Winter DeHon Problem Given: clocked circuit Goal: minimize clock period without changing (observable) behavior I.e. minimize maximum delay between any pair of registers Freedom: move placement of internal registers
CALTECH CS137 Winter DeHon Other Goals Minimize number of registers in circuit Achieve target cycle time Minimize number of registers while achieving target cycle time …start talking about minimizing cycle...
CALTECH CS137 Winter DeHon Simple Example Path Length (L) = 4 Can we do better?
CALTECH CS137 Winter DeHon Legal Register Moves Retiming Lag/Lead
CALTECH CS137 Winter DeHon Canonical Graph Representation Separate arc for each path Weight edges by number of registers (weight nodes by delay through node)
CALTECH CS137 Winter DeHon Critical Path Length Critical Path: Length of longest path of zero weight nodes Compute in O(|E|) time by levelizing network: Topological sort, push path lengths forward until find register.
CALTECH CS137 Winter DeHon Retiming Lag/Lead Retiming: Assign a lag to every vertex weight(e) = weight(e) + lag(head(e))-lag(tail(e))
CALTECH CS137 Winter DeHon Valid Retiming Retiming is valid as long as: – e in graph weight(e) = weight(e) + lag(head(e))-lag(tail(e)) 0 Assuming original circuit was a valid synchronous circuit, this guarantees: –non-negative register weights on all edges no travel backward in time :-) –all cycles have strictly positive register counts –propagation delay on each vertex is non-negative (assumed 1 for today)
CALTECH CS137 Winter DeHon Retiming Task Move registers assign lags to nodes –lags define all locally legal moves Preserving non-negative edge weights –(previous slide) –guarantees collection of lags remains consistent globally
CALTECH CS137 Winter DeHon Retiming Transformation N.B.: unchanged by retiming –number of registers around a cycle –delay along a cycle Cycle of length P must have –at least P/c registers on it –to be retimeable to cycle c
CALTECH CS137 Winter DeHon Optimal Retiming There is a retiming of –graph G –w/ clock cycle c –iff G-1/c has no cycles with negative edge weights G- subtract from each edge weight
CALTECH CS137 Winter DeHon 1/c Intuition Want to place a register every c delay units Each register adds one Each delay subtracts 1/c As long as remains more positives than negatives around all cycles –can move registers to accommodate –Captures the regs=P/c constraints
CALTECH CS137 Winter DeHon G-1/c
CALTECH CS137 Winter DeHon Compute Retiming Lag(v) = shortest path to I/O in G-1/c Compute shortest paths in O(|V||E|) –Bellman-Ford –also use to detect negative weight cycles when c too small
CALTECH CS137 Winter DeHon Bellman Ford For I 0 to N –u i (except u i =0 for IO) For k 0 to N –for e i,j E u i min(u i, u j +w(e i,j )) For e i,j E //still update negative cycle if u i >u j +w(e i,j ) –cycles detected
CALTECH CS137 Winter DeHon Apply to Example
CALTECH CS137 Winter DeHon Try c=1
CALTECH CS137 Winter DeHon Apply: Find Lags Negative weight cycles? Shortest paths?
CALTECH CS137 Winter DeHon Apply: Lags
CALTECH CS137 Winter DeHon Apply: Move Registers weight(e) = weight(e) + lag(head(e))-lag(tail(e))
CALTECH CS137 Winter DeHon Apply: Retimed
CALTECH CS137 Winter DeHon Apply: Retimed Design
CALTECH CS137 Winter DeHon Revise Example (fanout delay)
CALTECH CS137 Winter DeHon Revised: Graph
CALTECH CS137 Winter DeHon Revised: Graph
CALTECH CS137 Winter DeHon Revised: C=1?
CALTECH CS137 Winter DeHon Revised: C=2?
CALTECH CS137 Winter DeHon Revised: Lag
CALTECH CS137 Winter DeHon Revised: Lag Take ceiling to convert to integer lags: 0 0
CALTECH CS137 Winter DeHon Revised: Apply Lag 0 0
CALTECH CS137 Winter DeHon Revised: Apply Lag
CALTECH CS137 Winter DeHon Revised: Retimed
CALTECH CS137 Winter DeHon Pipelining We can use this retiming to pipeline Assume we have enough (infinite supply) registers at edge of circuit Retime them into circuit
CALTECH CS137 Winter DeHon C>1 ==> Pipeline
CALTECH CS137 Winter DeHon Add Registers Note: jump to G-1 graph future…show G first
CALTECH CS137 Winter DeHon Pipeline Retiming: Lag
CALTECH CS137 Winter DeHon Pipelined Retimed
CALTECH CS137 Winter DeHon Real Cycle
CALTECH CS137 Winter DeHon Real Cycle
CALTECH CS137 Winter DeHon Cycle C=1?
CALTECH CS137 Winter DeHon Cycle C=2?
CALTECH CS137 Winter DeHon Cycle: C-slow Cycle=c C-slow network has Cycle=1
CALTECH CS137 Winter DeHon 2-slow Cycle C=1
CALTECH CS137 Winter DeHon 2-Slow Lags
CALTECH CS137 Winter DeHon 2-Slow Retime
CALTECH CS137 Winter DeHon Retimed 2-Slow Cycle
CALTECH CS137 Winter DeHon C-Slow applicable? Available parallelism –solve C identical, independent problems e.g. process packets (blocks) separately e.g. independent regions in images Commutative operators –e.g. max example
CALTECH CS137 Winter DeHon Max Example
CALTECH CS137 Winter DeHon Max Example
CALTECH CS137 Winter DeHon Note Algorithm/examples shown –for special case of unit-delay nodes For general delay, –a bit more complicated –still polynomial
CALTECH CS137 Winter DeHon Initial State What about initial state? 0 1
CALTECH CS137 Winter DeHon Initial State 0
CALTECH CS137 Winter DeHon Initial State In general, constraints satisfiable?
CALTECH CS137 Winter DeHon Initial State ,1? 1 0
CALTECH CS137 Winter DeHon Initial State 1 0 Cycle1: 1 Cycle2: /(0*/in)=1 init=0 Cycle1: 1 Cycle2: /(/init*/in)=in init=1 Cycle1: 0 Cycle2: /(/init*/in)=1 ? Cycle1: /init Cycle2: /(/init*/in)=in+init init
CALTECH CS137 Winter DeHon Initial State Cannot always get exactly the same initial state behavior on the retimed circuit –without additional care in the retiming transformation –sometimes have to modify structure of retiming to preserve initial behavior Only a problem for startup transient –if you’re willing to clock to get into initial state, not a limitation
CALTECH CS137 Winter DeHon Minimize Registers
CALTECH CS137 Winter DeHon Minimize Registers Number of registers: w(e) After retime: w(e)+ (FI(v)-FO(v))lag(v) delta only in lags So want to minimize: (FI(v)-FO(v))lag(v) –subject to earlier constraints non-negative register weights, delays positive cycle counts
CALTECH CS137 Winter DeHon Minimize Registers Can be formulated as flow problem Can add cycle time constraints to flow problem Time: O(|V||E|log(|V|)log|(|V| 2 /|E|))
CALTECH CS137 Winter DeHon HSRA Retiming HSRA –adds mandatory pipelining to interconnect One additional twist –long, pipelined interconnect need more than one register on paths
CALTECH CS137 Winter DeHon Accommodating HSRA Interconnect Delays Add buffers to LUT LUT path to match interconnect register requirements Retime to C=1 as before Buffer chains force enough registers to cover interconnect delays
CALTECH CS137 Winter DeHon Accommodating HSRA Interconnect Delays
CALTECH CS137 Winter DeHon Summary Can move registers to minimize cycle time Formulate as a lag assignment to every node Optimally solve cycle time in O(|V||E|) time Also –Compute multithreaded computations –Minimize registers Watch out for initial values Can accommodate mandatory delays
CALTECH CS137 Winter DeHon Today’s Big Ideas Exploit freedom Formulate transformations (lag assignment) Express legality constraints Technique: –graph algorithms –network flow